Research On Optical Image Encryption Based On Deep Learning And Random Phase Encryption

With the rapid development of information technology,people are enjoying many conveniences while inevitably facing security risks such as information leakage and theft.Optical information security technology,as an important branch of information security,has received much attention due to its advantages of high parallelism,high security,and high processing speed.Meanwhile,deep learning technique has been widely applied in natural language processing and image recognition and has achieved a great success.Consequently,how to design and develop novel optical image encryption systems using deep learning technique has become one of the current research hotspots in the field of optical information security.This thesis first introduces research trends of optical information security and deep learning at home and abroad,and then describes the relevant theoretical basis,and finally introduces the main research work and the main achievements.The main research work of the thesis can be summarized as follows:1.An optical color image cryptosystem is proposed based on the interference principle and deep learning.This method first converts the color plaintext image into a color interferogram based on the interference principle.Then,the interferograms are Arnold transformed,and the resultant scrambled image is kept as the ciphertext.Subsequently,deep learning technique is employed to obtain the mapping relationship between the interferograms and plaintext by training a deep network model,which is later used as the key for decryption.In the decryption process,one can achieve a successfully decryption only with both the correct Arnold transformation times and network model parameters.Simulation results show that the proposed method has high robustness and security.2.An optical image encryption scheme is proposed based on deep learning and computer-generated binary hologram.Firstly,the plaintext is scrambled by using the Arnold transformation algorithm.The scrambled image is then used to generate an amplitude hologram based on optical diffraction and a neural network.Next,the amplitude hologram is binarized with maximum inter-class variance method.The resultant binary hologram is kept as ciphertext.In the decryption,the scrambled image is reconstructed from the binary hologram based on Fresnel diffraction,and the plaintext can be obtained by performing inverse Arnold transform on the reconstructed scramble image.Compared with traditional deep learning methods and iterative algorithms,our proposed method for amplitude hologram generation is faster and results in better reconstruction quality.Simulation results demonstrate the feasibility and effectiveness of the encryption scheme.